Abstract: For the problem of blind multi-user detection in the DS-CDMA system, an improved kernel independent component analysis (KICA) algorithm was proposed. First, fifth-Newton iteration method was introduced to the traditional FastICA algorithm, and then, a new kernel function-hybrid kernel function was used to solve the separation of nonlinear mixed signals. Based on the proposed algorithm, the multi-user signal detection was realized. At last, the proposed algorithm was compared with the traditional FastICA and KICA algorithms, and the simulation results showed that the improved-KICA algorithm had the smaller bit error rate and good convergent speed.